Accelerated Optimization of Curvilinearly Stiffened Panels using Deep Neural Networks

An important objective for the aerospace industry is to design robust and fuel efficient aerospace structures. The advanced manufacturing techniques like, additive manufacturing, have allowed structural designers to make use of curvilinear stiffeners for achieving better designs of stiffened plate and shell structures. The Finite Element Analysis (FEA) based standard optimization methods for aircraft panels with arbitrary curvilinear stiffeners are computationally expensive. The main reason for employing many of these standard optimization methods is the ease of their integration with the FEA. However, each optimization requires multiple computationally expensive FEA evaluations, making their use impractical at times. To accelerate optimization, we propose the use of Deep Neural Networks (DNNs) to approximate both the objective function and constraints, computed using MSC NASTRAN. The results are verified with those found in the literature. Later, the Python script was used to generate a large data-set using parallel processing. The 80%, 10% and 10% of the generated data-set are used for training, validation and testing of DNNs. We explored several techniques for improving the DNNs like logarithmic scaling, normalization, dropout regularization and different optimizer. The results show that DNNs obtained an accuracy of 95% on the test set for approximating FEA response within 10% of actual value. To compare the efficiency of the DNN, the wall-clock time is measured for running FEA and DNNs for evaluating buckling load of 200 aircraft panels. The DNN takes 10,000 times less wall-clock time than FEA. Our work demonstrates the potential of DNN-based machine learning algorithms for accelerating the optimization of stiffened panels.